Mapping Direct Seeded Rice in of ,

Murail Krishna Gumma, Deepika Uppala, Irshad A. Mohammed, Anthony M. Whitbread, and Ismail Rafi Mohammed

Abstract Across South Asia, the cost of rice cultivation has increased rice crop. Remote sensing technology provides a time saving due to labor shortage. Direct seeding of rice is widely pro- approach to estimate cropped area, intensity and other LULC moted in order to reduce labor demand during crop establish- changes in a country (Badhwar, 1984; Gumma et al., 2011a; ment stage, and to benefit poor farmers. To facilitate planning Gumma et al., 2015; Lobell et al., 2003; Thenkabail et al., and to track farming practice changes, this study presents 2009a; Thenkabail, 2010; Thiruvengadachari and Sakthiva- techniques to spatially distinguish between direct seeded and divel, 1997). Many studies have reported the use of spatial- transplanted rice fields using multiple-sensor remote sensing temporal data to map irrigated areas, land-use, land-cover, imagery. The District of Raichur, a major region in northeast and crop type (Dheeravath et al., 2010; Goetz et al., 2004; Karnataka, Central India, where irrigated rice is grown and Gumma et al., 2014; Knight et al., 2006; Thenkabail et al., direct seeded rice has been widely promoted since 2000, was 2005; Varlyguin et al., 2001; Velpuri et al., 2009) using MODIS selected as a case study. The extent of cropland was mapped NDVI time-series data to map both agricultural areas (Biggs et using Landsat-8, Moderate Resolution Imaging Spectroradi- al., 2006; Gaur et al., 2008; Gumma et al., 2011c) and seasonal ometer (MODIS) 16-day normalized difference vegetation index crop area (Sakamoto et al., 2005). There are also studies that (NDVI) time-series data and the cultivation practice delineated have used Synthetic Aperture Radar (SAR) data to monitor using RISAT-1 data for the year 2014. Areas grown to rice were rice areas, particularly irrigated rice. (Le Toan et al., 1997) mapped based on the length of the growing period detected used ERS-1 data as input to crop growth simulation models to using spectral characteristics and intensive field observa- estimate production for study sites in Indonesia and Japan. tions. The high resolution imagery of Landsat-8 was useful to Similarly, (Shao et al., 2001) mapped rice areas using tempo- classify the rice growing areas. The accuracy of land-use/land- ral RADARSAT data (1996 and 1997) for production estimates cover (LULC) classes varied from 84 percent to 98 percent. The in Zhaoqing in China. (Imhoff and Gesch, 1990) derived results clearly demonstrated the usefulness of multiple-sensor sub-canopy digital terrain models of flooded forest usingSAR . imagery from MOD09Q1, Landsat-8, and RISAT-1 in mappingDelivered the by Ingenta(Kasischke and Bourgeau-Chavez, 1997) monitored the wet- rice area and practices accurately, IP:routinely, 192.168.39.211 and consistently. On: Mon, lands27 Sep of South2021 21:14:55Florida using ERS-1 SAR imagery. (Leckie, 1990) The low cost of imagery backedCopyright: by ground American survey, Society as dem for- Photogrammetrydistinguished and forest Remote type usingSensing SAR and optical data. (Robin- onstrated in this paper, can also be used across rice growing son et al., 2000) delineated drainage flow directions using SAR countries to identify different rice systems. data. (Townsend, 2001) mapped seasonal flooding in forested wetlands using temporal RADARSAT SAR imagery. (Bouvet and Le Toan, 2011) demonstrated how lower resolution wide- Introduction swath images from advanced SAR (ASAR) data could be used to Agriculture is an important sector in India, contributing about map rice over larger areas. (Uppala et al., 2015) mapped rice 17.9 percent of the gross domestic product (GDP) (2014 figures) areas using single data hybrid polarimetric data from RISAT-1 (CIA, 2015). About 68 percent of the country’s population lives SAR data. The advantage of rice mapping with SAR is that it in rural areas (FAOSTAT, 2013). India is one of the largest rice can overcome pervasive cloud cover across Asia during the growing nations accounting for one quarter of the global rice rainy season months when rice is cultivated. However, its area and produces around 125 million tons/year, with low high cost inhibits its large-scale application. yields of around 2.85 t/ha (Siddiq, 2000). Given that rice is a A comprehensive plan to increase agricultural productivity staple food crop in India and that the country has a popula- and manage labor can be drawn up based on near real-time tion that is growing at 1.2 percent per annum, a substantial monitoring of rice systems along with weather and crop data increase in rice production is the only way to guarantee food acquired from an analysis of remote sensing imagery. Remote security and to keep food prices low. Labor and fertilizer are sensing platforms have proved to be ideal to take resource the two major input costs that are incurred while growing rice inventories and monitor agriculture. Interdisciplinary ap- when there is assured water availability (Savary et al., 2005). proaches and methods have made it easier to analyze imagery Migration from rural to urban areas in search of work has led and extract information in ways unknown a few decades ago. to labor shortages and higher costs of agricultural operations The goal of this study was to develop a method using mul- (Hossain, 1998; Savary et al., 2005). To overcome these con- tiple-sensor imagery (Landsat-8, RISAT-1 and MODIS time-series straints, promoting direct seeding in lieu of the traditional and data) to map the spatial extent of two different rice cultiva- labor-intensive transplanting (De Datta, 1986; Naylor, 1992) is tion practices (direct seeding and transplanting) in Raichur being seen as a feasible alternative. However, there is a trad- district, Karnataka State, India. eoff with lower yields in rice that is direct seeded rather than transplanted; so efforts are on to bridge this gap (Khush, 1995). Accurate and timely information on rice systems, cropped area, and yields are very important for sustainable food secu- Photogrammetric Engineering & Remote Sensing rity. Several studies have used remote sensing data to monitor Vol. 81, No. 11, November 2015, pp. 873–880. 0099-1112/15/873–880 International Crops Research Institute for the Semi-Arid © 2015 American Society for Photogrammetry Tropics (ICRISAT), Patancheru, Telangana, India (m.gumma@ and Remote Sensing cgiar.org). doi: 10.14358/PERS.81.11.873

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING November 2015 873

11-15 November Peer Reviewed.indd 873 10/21/2015 1:18:30 PM Study Area and Data Sets Ground Survey Data A ground survey was conducted during the monsoon season Study Area for the crop year 2014 to 2015. Data was collected at 238 loca- Raichur is one of largest and most productive rice growing tions covering the major cropland areas in the study area. All district in Karnataka state in southern India (Figure 1). The location-specific data were collected from 250 × 250 meter district lies between 15°42'57" and 16°26'39" latitude and plots and consisted of location coordinates (latitude, longi- 76°14'47" and 77°35'39" longitude, with a land mass of 835,843 tude), land-use categories, land-use (percent), cropping systems ha (cropped area 695,000 ha) (Figure 1) and bound in the during the monsoon season (through farmer interviews), crop north and south by the rivers Krishna and Tungabhadra. Agro- types, and watering method (irrigated/rainfed). Samples were climatically, it falls under the Northeast Dry Zone of Karnataka within large contiguous areas of a specific land-use/land-cover. (semi-arid eco-subregion) with a long term average rainfall of The locations were chosen based on pre-classification classes just above 600 mm distributed over kharif and rabi seasons. and local expertise. The local experts also provided information Much of the district is well irrigated by the Tungabhadra Dam on cropping systems, cropping patterns and cropping inten- on the , and the Narayanpura Dam on the sity (single or double crop), irrigation application, and canopy Krishna River, covering an irrigated area of 185,000 ha, 72.2 cover ( percent) for the previous years for these locations from percent of which is from canals. The arable rainfed area occu- their recorded data. Overall, 116 spatially well-distributed data pies 405,000 ha. The major crops grown in the district include points (Figure 1) were used for class identification and labeling; paddy, sorghum, groundnut, sunflower, cotton, and pulses of these, 46 data points were used for ideal spectra generation (chickpea and pigeonpea). The rice-rice cropping system is and an additional 122 were used to assess accuracy. dominant in most irrigated areas. The soils of the district are mainly deep black clayey (47 percent) and red (34 percent). Table 1. Characteristics of Multiple Satellite Sensor Data Used in the Study Satellite Images Spatial Band range Four Landsat-8 tiles were downloaded from the Earth Ex- Sensor Dates Bands (m) (µm) plorer, global land cover facility website (http://earthexplorer. usgs.gov/). Images from the 2014 monsoon season and their 1 0.433–0.453 spectral characteristics are shown in Table 1. All the Landsat-8 2 0.450–0.515 tiles were converted into reflectance http://landsat.usgs.gov/( documents/Landsat8DataUsersHandbook.pdf) to normalize 3 0.525–0.600 07 & 14 the multi-date effect (Markham and Barker, 1986; Thenkabail Landsat-8 30 4 0.630–0.680 et al., 2004) using the spatial modeler in ERDAS Imagine® (ER- Nov 2014 5 0.845–0.885 DAS, 2007). The 250-m, two-band MODIS data (centered at 648 nm and 858 nm; Table 1) collection 5 (MOD09Q1) were acquired 6 1.560–1.660 for every eight days during the crop-growing seasons from 7 2.100–2.300 January 2014 through December 2014. The data was acquiredDelivered by Ingenta in 12-bit (0 to 4,096 levels) and stretchedIP: to192.168.39.211 16-bit (0 to 65,536 On: Mon, 27 Sep 202128 Jun 21:14:55 2014; levels). Further processing stepsCopyright: are described American in (Gumma Society et for Photogrammetry23 and Jul 2014; Remote Sensing RISAT-1 17 Aug 2014; 18 1 HH polarization al., 2015; Gumma et al., 2011b; Thenkabail et al., 2005). 11Sep 2014; RISAT-1, C-band (5.35 GHz) HH polarization, Medium Resolu- 06 Oct 2014 tion SAR (MRS) temporal data sets were used to distinguish direct seeded rice from transplanted rice. Data was acquired 1 0.62–0.67 MODIS 2014 every 25 days at a 37° angle of incidence spaced phenologi- 250 2 0.84–0.88 cally appropriate to identify rice fields and provide informa- (MOD09Q1) (16 days) tion on data sets (Table 1). NDVI - 1 to + 1

Figure 1. Raichur District, Karnataka, India.

874 November 2015 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING

11-15 November Peer Reviewed.indd 874 10/21/2015 1:18:30 PM Methods classification algorithm (in ERDAS Imagine 2010) was applied Figure 2 presents an overview of the comprehensive meth- on a 12-band NDVI (monthly Maximum Value Composite) odology used to map direct seeded and transplanted rice MVC to obtain the initial 100 classes, followed by progressive cultivation practices using time series imagery from MOD09Q1 generalization (Cihlar et al., 1998). The unsupervised classifi- (spatial resolution 250 m) 16-day time-series NDVI, Landsat-8 cation was set at a maximum of 100 iterations with a conver- (spatial resolution 30 m), and RISAT-1 (spatial resolution 18 m). gence threshold of 0.99 (Leica, 2010). Time-series NDVI spectra Rice growing areas were delineated initially with MODIS time- were then plotted for each of the 100 classes and compared series NDVI using spectral matching techniques (Gumma et with the ideal spectra to identify and label classes (Gumma et al., 2011b; Gumma et al., 2014; Thenkabail et al., 2007a). This al., 2014). However, the time-series NDVI profile helps gain an was used as a mask to clip Landsat-8 time-series imagery. The understanding of the growth profile of different crops in addi- subset of Landsat-8 was reclassified to separate rice from non- tion to providing information on planting date, discrimination rice areas due to the difference in resolution between MODIS between rice and other crops, early stage conditions (flooded and Landsat-8. The Landsat-8 output provided a more ac- pixel showing low values initially), and discrimination be- curate delineation of rice growing area. This subset was again tween irrigation sources (e.g., irrigated versus rainfed). Class used as a mask to clip out RISAT-1 temporal imagery. identification and labeling were performed based on a suite of methods and ancillary data, such as decision tree algorithms, Land-Use / Land-Cover Classification spectral matching techniques, Google Earth™ high-resolution The procedure began with image normalization of Landsat-8 imagery and ground survey data (Gumma et al., 2014; Thenka- data converted to top of atmosphere (TOA) reflectance using a bail et al., 2009b; Thenkabail et al., 2007b). The initial reduc- reflectance model implemented in ERDAS Imagine http://land( - tion in classes used a decision tree method (De Fries et al., sat.usgs.gov/documents/Landsat8DataUsersHandbook.pdf). 1998) based on the temporal NDVI data. The decision tree is The (Operational Land Imager) OLI band data can be con- based on NDVI thresholds at different stages in the season that verted to TOA planetary reflectance using Reflectance rescaling define vegetation growth cycle, and these algorithms help to coefficients provided in the product metadata file. The follow- identify similar classes. The dates and threshold values were ing equation was used to convert DN values to TOA planetary derived from the ideal temporal profile (Gummaet al., 2014). reflectance forOLI data: Using the ground survey data, Google Earth’s high-resolution imagery along with spectral profiles of rice crops fromMODIS ρλ′ = Mρ Qcal + Aρ (1) imagery, Landsat-8 imagery was classified using the super- where: ρλ′ = TOA planetary reflectance (without correction of vised maximum likelihood classification algorithm. The MODIS-derived rice area was used as the basis of the solar angle), Mρ = Band specific multiplicative rescaling factor maximum possible extent of rice area as one segment and from the metadata, Aρ = Band specific additive rescaling factor LULC from the metadata, and Qcal = the quantized and calibrated other areas as another segment. This was used as a mask standard product pixel values (DN). to clip Landsat-8 time-series imagery. The subset of Landsat-8 TOA reflectance with correction for the sun angleDelivered is then: by Ingentawas classified again to separate rice from non-rice areas. Both IP: 192.168.39.211 On: Mon, segments27 Sep 2021 were 21:14:55classified independently (to avoid mixed clas- Copyright:ρλ' American Society for Photogrammetry and Remote Sensing ρλ = (2) sification) using the protocols mentioned earlier. This led to sin(θSE ) a more accurate delineation of rice growing area, but did not show fragmented direct seeded rice areas. This was mainly where: ρλ = TOA planetary reflectance,ρλ ′ = TOA planetary because direct seeded rice was sown in the early monsoon, reflectance (without correction of solar angle), andθ SE = the when there were no images due to heavy clouds. Landsat-8 local sun elevation angle provided in the metadata. rice area was again used as a mask to clip out RISAT-1 temporal The MODIS stacked composite was classified using unsuper- imagery. Separating the two practices of rice cultivation was vised ISOCLASS cluster K-means classification algorithm fol- possible using RISAT-1 temporal imagery. This is the best com- lowed by successive generalization (Biggs et al., 2006; Gumma plimentary data to monitor croplands during the monsoon et al., 2011c; Thenkabail et al., 2005). The unsupervised season, and where there are continuous clouds.

Figure 2. Overview of the methodology for mapping different rice growing practices.

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING November 2015 875

11-15 November Peer Reviewed.indd 875 10/21/2015 1:18:36 PM Delineating Direct Seeded Rice (DSR) Cultivation Using SAR Data The RISAT-1 data was processed using ENVI software. The lev- el-2, HH polarization data was imported in ENVI raster format (.hdr). To reduce noise, Enhanced Lee Adaptive Speckle Filter with kernel size of 5 × 5 was applied (Lopes et al., 1990). Radiometric calibration for HH polarization amplitude image was carried out by using the metadata provided along with the imagery. The scaled values of digital numbers were converted into backscattering coefficient using the equation given in (Chakraborty et al., 2013; Laur et al., 2002). The temporal im- ages were co-registered by fitting a second order polynomial model to the manual GCPs with 50 well distributed points throughout the scene. The temporal HH polarization RISAT-1 data was stacked. The rice crop delineated from Landsat-8 was used to subset the RISAT-1 data stack for further processing. In Raichur, direct seeded rice cropping is practiced along with transplanted rice which is most prevalent. Transplanted rice is of two types depending on sowing date. The early transplanted Figure 3. Temporal changes in backscatter coefficient in different rice is sown in August and the late transplanted rice in Septem- rice growing practices. ber. SAR data is unique since it can distinguish between different rice sowing dates. The rice crop generates a unique temporal backscatter profile. It is clear from Figure 3 that during the estab- distinguished at the transplantation stage due to very low lishment stage of transplanted rice, the backscatter is higher than backscatter. However, it is difficult to discriminate between direct seeded rice due to tilling and low moisture. During trans- direct seeded rice and late transplantation due to their similar plantation, backscatter is quite low due to standing water which growth stages. The deviation of backscatter was found to be reflects minimal energy in the backscatter direction. Further, low. The decision tree hybrid classification algorithm was the backscatter increases as the plant grows because of multiple used here in order to separate the above mixed classes. June reflections from the crop canopy (Choudhuryet al., 2012). is the crucial month to separate transplanted rice form direct In the establishment phase of direct seeded rice, low seeded rice (80 percent). The early transplanted rice was eas- backscatter was observed due to moisture conditions. This ily separated from directed seeded rice by applying decision is the key to distinguishing 80 percent of the direct seeded rules on the August imagery. This was possible because the rice from transplanted fields. Also, some of the transplanted difference in backscatter was more than 2 dB. The late trans- rice fields cannot be distinguished from direct seeded fields plantation was not separable from direct seeded rice because due to high standard deviation in transplanted fields becauseDelivered ofby the Ingenta low difference in backscatter. In order to resolve these of roughness. Additionally, early transplantedIP: 192.168.39.211 rice could be On: Mon,types 27 of Sep fields, 2021 unsupervised 21:14:55 k-means algorithm with a con- Copyright: American Society for Photogrammetryvergence threshold and of Remote 0.99 was Sensing applied on the remaining area

Plate 1. Land-use/land-cover during the 2014 monsoon season (kharif) in Raichur District, with rice classes temporal NDVI profiles.

876 November 2015 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING

11-15 November Peer Reviewed.indd 876 10/21/2015 1:18:39 PM which included late transplanted rice and direct seeded rice. irrigation and estimate harvest dates for each type of practice. By using this algorithm, 85 percent of late transplanted rice It is clear that the head end of the canal has a larger extent fields were separated from direct seeded rice fields. of transplanted rice than the tail end in and Sindha- nur with irrigated mixed crops. Rainfed croplands occupy the largest area of 351,895 ha in Raichur District. Pigeonpea Results mixed with a cereal crop like sorghum is one of the important and large cropping systems in the taluks of Lingsugur and Spatial Distribution of Land Use / Land Cover Raichur. Rangelands and shrublands are comparably large Ten land-use/land-cover classes were identified using spectral which can be potentially used for rainfed agriculture. matching techniques using MODIS temporal imagery and Land- sat-8 (Plate 1). With average rainfall in the order of 600 mm/ Spatial Extent of Direct Seeded Rice and Transplanted Rice annum and large areas under rainfed cropping, canal irriga- Rice is a major crop in Raichur District, covering 130,343 ha in tion from Tungabhadra Dam provides insufficient irrigation 2013, out of which transplanted rice covered 117,401 ha and to grow rice. The total irrigated area accounts for 287,993 ha direct seeded rice 12,942 ha. The major rice area is located in (Table 2). The important rice growing taluks (administrative Manvi and taluks which are irrigated, depending on units) are Manvi and Sindhanur where a large area is irri- ground water availability and rainfall. Of the total rice area, 52 gated under rice. The adoption of direct rice seeding is due to percent is early transplanted rice, 38 percent is late transplant- shortage of labor, and it is a new phenomenon and expanding ed rice, and only 10 percent is direct seeded rice (Table 2 and along with transplanted rice cultivation which is essentially Plate 1). Variations within the direct seeded field are difficult large in extent. The use of temporal RISAT-1 imagery helped to identify due to the temporal deficit of the image acquisition. in identifying how rice was planted (direct seeded rice or transplanted) and also ascertaining time of planting (early or Accuracy Assessment late transplanting). This information can be used to schedule The classification accuracies obtained from the 122 indepen- dent ground data observation data points for the year 2014 Table 2. Estimated Area Under Different Land Use/Land Cover (LULC) were gathered from the study area (monsoon season). Accuracy Classes in Raichur District assessment was done using error matrix, a multi-dimensional LULC area Area (ha) % table in which the cells contain changes from one land-use class to another. The statistical approach of accuracy assess- 01. Irrigated-conjunctive-DS-rice 12,942 1.5 ment consists of different multi-variate statistical analyses 02. Irrigated-conjunctive-early TP-rice 67,418 8.0 (Congalton and Green, 1999; Jensen, 1996). The error matrix ac- curacy varied from 75 percent to 92 percent across six cropland 03. Irrigated-SW-late TP-rice 49,983 5.9 areas. However, it must be noted that rice classes did not mix with other classes. Overall, accuracy was 82 percent and kappa 04. Irrigated-conjunctive-SC-mixed crops 157,650 18.7 was 0.78. So the uncertainty of about 18 percent is due to the 05. Rainfed-SC-pigeonpea/mixed crops 122,778Delivered14.6 by Ingentainter-mix among the other land-use/land-cover classes (Table 3). IP: 192.168.39.211 On: Mon, 27 Sep 2021 21:14:55 06. Rainfed-mixed crops/fallowsCopyright: American 229,117Society 27.2for Photogrammetry and Remote Sensing 07. Rangelands/shrublands 152,442 18.1 Comparison with National Statistics and Other Studies Statistics on taluk-wise rice area in Raichur District was ob- 08. Barren lands/shrublands/forests 37,060 4.4 tained from the Bureau of Economics and Statistics, Govern- 09. Water bodies 9,241 1.1 ment of Karnataka State, India. This data was used for com- paring RISAT-1, Landsat-8, and MODIS-derived statistics. A high 10. Settlements 3,695 0.4 correlation was found (Figure 4) between the two sources of Total 842,326 information on rice cultivation in the district.

Table 3. Accuracy Assessment Using Ground Survey Data (Two Dimensional Error Matrix) Reference data

Classified data 01. Irrigated-conjuctive- DS-rice 02. Irrigated-conjunctive- early TP-rice 03. Irrigated-SW-late TP-rice 04. Irrigated-conjunctive- SC-mix 05. Rainfed-SC- pigeonpea / mixed crops 06. Rainfed-mixed crops/ fallows 07. Other LULC Reference row total Classified totals Producers accuracy Users accuracy Kappa accuracy 01. Irrigated-conjuctive-DS-rice 3 0 1 0 0 0 0 3 4 100% 75% 74% 02. Irrigated-conjuctive-early TP-rice 0 12 1 0 0 0 0 13 13 92% 92% 91% 03. Irrigated-SW-late TP-rice 0 0 7 1 0 0 0 9 8 78% 88% 87% 04. Irrigated-conjuctive-SC-mix 0 0 0 22 2 1 0 24 25 92% 88% 85% 05. Rainfed-SC-pigeonpea/mixed crops 0 0 0 1 19 2 1 27 23 70% 83% 78% 06. Rainfed-mixed crops/fallows 0 1 0 0 4 30 1 37 36 81% 83% 76% 07. Other LULC 0 0 0 0 2 4 7 9 13 78% 54% 50% Total 3 13 9 24 27 37 9 122 122 84% 80% 77%

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING November 2015 877

11-15 November Peer Reviewed.indd 877 10/21/2015 1:18:39 PM Discussion SAR data along with optical data such as MODIS, a similar Microwave remote sensing satellites carrying Synthetic Aper- approach can be used to map direct seeded rice cultivation ture RADAR (SAR) enable imaging of the features of the Earth’s during post-rainy (rabi) season as well. The spatial distribu- surface both during the day and the night under all weather tion and quantification of the extent of direct seeded rice conditions. These specific characteristics of the satellite are will not only help breeders to understand the locale-specific useful in identifying crop information under cloudy skies constraints to yields such as weeds, but also provide remedies using standard methods. However, the objective stated in to bridge the yield gaps. In turn, this will help in overcoming this study demands a hybrid approach where optical and SAR the labor shortage and also help in conserving water for other imagery are used to distinguish between the two rice growing uses. In this context, it is proposed to study and examine how practices that exhibit similar backscatter mechanisms, except we can differentiate between rice establishment practices not at the establishment stage. Studies have been conducted to only at the time of establishment using SAR imagery, but also come up with a robust classification method using X-band when there is a large outbreak of weeds in direct seeded rice SAR data to map rice crops, including for direct seeding in cultivation. different ecosystems and also to examine the relationship be- tween crop characteristics and backscatter coefficient (Nelson et al., 2014). However, the inherent property of C-band SAR Acknowledgments data, where the wavelength responds to roughness of soil by This research was supported by the CGIAR Research Program including any background (soil backscatter), makes a differ- Water, Land and Ecosystems. The authors would like to thank ence by identifying a practice like direct seeded rice. The Smitha Sitaraman and Amit Chakravarty, science editors/pub- basic premise that direct seeded rice requires less labor and lisher at ICRISAT for his assistance with editing. The authors less water, making it economically less demanding option also thank Dr. AN Rao (IRRI) and Dr. Gajanan Sawargaonkar and also a strategy to cope with monsoon failure, has some for their valuable feedback of the rice classification system lacunae. This is because the direct seeded rice fields are more and sub-district wise statistics. infested by weeds than transplanted rice fields (Raoet al., 2007) as the puddled fields do not allow weed growth. This leads to yield losses depending on weed infestation. References RISAT-1 SAR imagery was a viable and useful option as it Badhwar, G.D., 1984. Automatic corn-soybean classification using overcomes cloud cover as well as provides a convenient landsat MSS data - I. Near-harvest crop proportion estimation, temporal coverage to distinguish between these two practices Remote Sensing of Environment, 14(1-3):15–29. right at establishment stage. Mixed classes were resolved Biggs, T.W., P.S. Thenkabail, M.K. Gumma, C.A. Scott, G.R. based on extensive ground information collected during field Parthasaradhi, and H.N. Turral, 2006. Irrigated area mapping in ™ heterogeneous landscapes with MODIS time series, ground truth trips and tools like Google Earth Explorer . A multi-sensor and census data, Krishna Basin, India, International Journal of approach was a useful strategy to fulfill the objective of this Remote Sensing, 27(19):4245–4266. Delivered by Ingenta study, especially to arrive at accurate estimates of the prac- Bouvet, A., and T. Le Toan, 2011. Use of ENVISAT/ASAR wide-swath tice of growing direct seeded rice in RaichurIP: 192.168.39.211 District. The On: Mon, 27data Sep for timely 2021 rice 21:14:55 fields mapping in the Mekong River Delta, MODIS, 16-day NDVI data wasCopyright: primarily used American to delineate Society the for PhotogrammetryRemote Sensing and of Remote Environment Sensing, 115(4):1090–1101. rice cropped areas taking advantage of its temporal availabil- Chakraborty, M., S. Panigrahy, A. Rajawat, R. Kumar, T. Murthy, D. ity. The spectral profiles fromMODIS temporal data provide Haldar, A. Chakraborty, T. important information on different aspects of the crop, such Kumar, S. Rode, and H. Kumar, 2013. Initial results using RISAT-1 as its source of water, phenological stage, and stress. The role C-band SAR data, Current Science (), 104(4):490–501. of Landsat-8 imagery was to generate a high accuracy rice Choudhury, I., M. Chakraborty, S.C. Santra, and J.S. Parihar, 2012. cropped area map to match the resolution of RISAT-1 imagery Methodology to classify rice cultural types based on water and in turn transfer information from MODIS to RISAT-1. regimes using multi-temporal RADARSAT-1 data, International Karnataka State’s sub-district divisions are large and few. Journal of Remote Sensing, 33(13):4135–4160. For instance, Raichur has only five taluks. Statistics gener- CIA, 2015. The World Factbook, URL: https://www.cia.gov/library/ ated from this study on taluk-wise rice cropped area were publications/the-world-factbook/fields/2012.html (last date compared with those from the government department. A accessed: 24 September 2015). high correlation was observed between these two sources of Cihlar, J., Q. Xiao, J. Beaubien, K. Fung, and R. Latifovic, 1998. information (Figure 4). Statistics on direct seeded rice was not Classification by progressive generalization: A new automated available from the government departments since it is a new methodology for remote sensing multichannel data, International practice in this district. However, this study was successful Journal of Remote Sensing, 19(14):2685–2704. in identifying direct seeded rice because of the difference in Congalton, R.G., and K. Green, 1999. Assessing the Accuracy of crop establishment dates compared to the other practice of Remotely Sensed Data: Principles and Practices, New York: transplanting rice. Lewis. De Datta, S., 1986. Technology development and the spread of direct- seeded flooded rice in Southeast Asia,Experimental Agriculture, Conclusions 22(04):417–426. Mapping the cropped area of direct seeded rice using a multi- De Fries, R.S., M. Hansen, J.R.G. Townshend, amd R. Sohlberg, 1998. Global land cover classifications at 8 km spatial resolution: The sensor approach, specifically the use ofRISAT-1 imagery, is use of training data derived from Landsat imagery in decision important as it can penetrate cloud cover during the rainy tree classifiers,International Journal of Remote Sensing, season. Optical remote sensing methods may not be success- 19(16):3141–3168. ful in discriminating land-use classes based on management Dheeravath, V., P.S. Thenkabail, G. Chandrakantha, P. Noojipady, practices. Hybrid approaches have always proved useful in G.P.O. Reddy, C.M. Biradar, M.K. Gumma, and M. Velpuri, 2010. atypical cases such as this study. Even though there are three Irrigated areas of India derived using MODIS 500 m time series types of direct seeded rice cultivation, an attempt was made for the years 2001-2003, ISPRS Journal of Photogrammetry and to map only the dry seeded sowing practice prevalent in Remote Sensing, 65(1):42–59. Raichur District. Sindhanur and Manvi are the two taluks in ERDAS, 2007. ERDAS Field Guide, Volume 1, October 2007. the district where the practice is followed. Using the temporal

878 November 2015 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING

11-15 November Peer Reviewed.indd 878 10/21/2015 1:18:39 PM FAOSTAT, 2013. URL: http://faostat.fao.org/site/666/default.aspx (last Lobell, D.B., G.P. Asner, J.I. Ortiz-Monasterio, and T.L. Benning, 2003. date accessed: 24 September 2015). Remote sensing of regional crop production in the Yaqui Valley, Gaur, A., T.W. Biggs, M.K. Gumma, and H. Turral, 2008. Water Mexico: Estimates and uncertainties, Agriculture, Ecosystems & scarcity effects on equitable water distribution and land use Environment 94(2):205–220. in Major Irrigation Project - A Case study in India, Journal of Lopes, A., R. Touzi, and E. Nezry, 1990. Adaptive speckle filters Irrigation and Drainage Engineering, 134 (1):26–35. and scene heterogeneity, IEEE Transactions on Geoscience and Goetz, S.J., D. Varlyguin, A.J. Smith, R.K. Wright, S.D. Prince, Remote Sensing, 28(6):992–1000. M.E. Mazzacato, J. Tringe, C. Jantz, and B. Melchoir, 2004. Markham, B.L., and J.L. Barker, 1986. Landsat MSS and TM Post - Application of multitemporal Landsat data to map and monitor Calibration Dynamic Ranges, Exoatmospheric Reflectances and land cover and land use change in the Chesapeake Bay At-Satellite Temperatures, Earth Observation Satellite Company, watershed, Analysis of Multi-temporal Remote Sensing Images Lanham, Maryland, Landsat Technical Notes, 1, August. (P. Smits and L. Bruzzone, editors), World Scientific Publishers, Naylor, R., 1992. Labour-saving technologies in the Javanese rice Singapore, pp. 223–232. economy: Recent developments and a look into the 1990s, Gumma, M.K., S. Mohanty, A. Nelson, R. Arnel, I.A. Mohammed, Bulletin of Indonesian Economic Studies, 28(3):71–91. and S.R. Das, 2015. Remote sensing based change analysis of Nelson, A., T. Setiyono, A.B. Rala, E.D. Quicho, J.V. Raviz, P.J. rice environments in Odisha, India, Journal of Environmental Abonete, A.A. Maunahan, C.A. Garcia, H.Z.M. Bhatti, and Management, 148(0):31–41. L.S. Villano, 2014. Towards an operational SAR-based rice Gumma, M.K., P.S. Thenkabail, A. Maunahan, S. Islam, and A. monitoring system in Asia: Examples from 13 demonstration Nelson, 2014. Mapping seasonal rice cropland extent and area sites across Asia in the RIICE Project, Remote Sensing, in the high cropping intensity environment of Bangladesh 6(11):10773–10812. using MODIS 500m data for the year 2010, ISPRS Journal of Rao, A.N., D.E. Johnson, B. Sivaprasad, J.K. Ladha, and A.M. Photogrammetry and Remote Sensing, 91(5):98–113. Mortimer, 2007. Weed management in direct‐seeded rice, Gumma, M.K., D. Gauchan, A. Nelson, S. Pandey, and A. Rala, 2011a. Advances in Agronomy (L.S.Donald, editor), Academic Press, Temporal changes in rice-growing area and their impact on pp. 153–255. livelihood over a decade: A case study of Nepal, Agriculture, Robinson, C., F. El-Baz, M. Ozdogan, M. Ledwith, D. Blanco, S. Ecosystems & Environment, 142(3-4):382–392. Oakley, and J. Inzana, 2000. Gumma, M.K., A. Nelson, P.S. Thenkabail, and A.N. Singh, 2011b. Use of radar data to delineate palaeodrainage flow directions in Mapping rice areas of South Asia using MODIS multitemporal the Selima Sand Sheet, Eastern Sahara, Photogrammetric data, Journal of Applied Remote Sensing, 5, 053547. Engineering & Remote Sensing, 66(6):745–753. Gumma, M.K., P.S. Thenkabail, I.V. Muralikrishna, M.N. Velpuri, P.T. Sakamoto, T., M. Yokozawa, H. Toritani, M. Shibayama, N. Ishitsuka, Gangadhararao, V. Dheeravath, C.M. Biradar, S. Acharya Nalan, and H. Ohno, 2005. A crop phenology detection method using and A. Gaur, 2011c. Changes in agricultural cropland areas time-series MODIS data, Remote Sensing of Environment, 96(3- between a water-surplus year and a water-deficit year impacting 4):366–374. food security, determined using MODIS 250 m time-series data Savary, S., N.P. Castilla, F. Elazegui, and P.S. Teng, 2005. Multiple and spectral matching techniques, in the Krishna River basin effects of two drivers of agricultural change, labour shortage and (India), , 32(12):3495– International Journal of Remote Sensing Delivered by Ingentawater scarcity, on rice pest profiles in tropical Asia,Field Crops 3520. IP: 192.168.39.211 On: Mon, 27 SepResearch 2021, 91(2):263–271. 21:14:55 Hossain, M., 1998. SustainingCopyright: Food Security American in Asia: SocietyEconomic, for PhotogrammetryShao, Y., X. Fan, and H. Remote Liu, J. Xiao, Sensing S. Ross, B. Brisco, R. Brown, and G. Social, and Political Aspects - Sustainability of Rice in the Staples, 2001. Rice monitoring and production estimation using , Pacific Basin Study Center, Davis, Global Food System multitemporal RADARSAT, Remote Sensing of Environment, California, and International Rice Research Institute, Manila, 76(3):310–325. The Philippines, pp. 19–44. Siddiq, E., 2000. Bridging the Rice Yield Gap in India, URL: http:// Imhoff, M.L., and D.B. Gesch, 1990. The derivation of a sub-canopy www.fao.org/docrep/003/X6905E/x6905e09.htm (last date digital terrain model of a flooded forest using synthetic aperture accessed: 24 September 2015). radar, Photogrammetric Engineering & Remote Sensing, 56(8):1155–1162. Thenkabail, P.S., 2010. Global croplands and their importance for water and food security in the twenty-first century: Towards an Jensen, J.R., 1996. Introductory Digital Image Processing: A Remote ever green revolution that combines a second green revolution , Prentice Hall, Upper Saddle River, New Sensing Perspective with a blue revolution, Remote Sensing, 2(9):2305–2312. Jersey. Thenkabail, P., C. Biradar, P. Noojipady, V. Dheeravath, Y. Li, M. Kasischke, E.S., and L.L. Bourgeau-Chavez, 1997. Monitoring south Velpuri, M. Gumma, G. Florida wetlands using ERS-1 SAR imagery, Photogrammetric Engineering & Remote Sensing 63(3):281–291. Reddy, H. Turral, X. Cai, J. Vithanage, M. Schull, and R. Dutta, 2009a. Global irrigated area map (GIAM) for the end of the last Khush, G.S., 1995. Breaking the yield frontier of rice, , GeoJournal millennium derived from remote sensing, International Journal 35(3):329–332. of Remote Sensing, 30(14):3679–3733. Knight, J.F., R.L. Lunetta, J. Ediriwickrema, and S. Khorram, 2006. Thenkabail, P.S., C.M. Biradar, P. Noojipady, V. Dheeravath, Y. Li, M. Regional scale land-cover characterization using MODIS-NDVI Velpuri, M., Gumma, O.R.P. Gangalakunta, H. Turral, X. Cai, J. 250 m multi-temporal imagery: A phenology based approach, Vithanage, M.A. Schull, and R. Dutta, 2009b. Global irrigated GIScience and Remote Sensing, 43(1):1–23. area map (GIAM), derived from remote sensing, for the end of Laur, H., P. Bally, P., Meadows, J. Sanchez, B. Schaettler, E. Lopinto, the last millennium, International Journal of Remote Sensing, and D. Esteban, 2002. 30(14):3679–3733. Derivation of the backscattering coefficient σ0 in ESA ERS SAR PRI Thenkabail, P., P. GangadharaRao, T. Biggs, M. Gumma, and H. Turral, products, ESA, Noordjiwk, The Netherlands, ESA Document ES- 2007a. Spectral matching techniques to determine historical TN-RE-PM-HL09(2). land-use/land-cover (LULC) and irrigated areas using time- Le Toan, T., F. Ribbes, L.-F. Wang, N. Floury, K.-H. Ding, J.A. Kong, M. series 0.1-degree AVHRR Pathfinder datasets,Photogrammetric Fujita, and T. Kurosu, 1997. Rice crop mapping and monitoring Engineering & Remote Sensing, 73(10):1029–1040. using ERS-1 data based on experiment and modeling results, IEEE Thenkabail, P.S., P. GangadharaRao, Y. Biggs, M.K. Gumma, and Transactions on Geoscience and Remote Sensing, 35(1):41–56. H. Turral, 2007b. Spectral Matching techniques to determine Leckie, D., 1990. Synergism of synthetic aperture radar and visible/ historical land use/land cover (LULC) and irrigated areas using infrared data for forest type discrimination, Photogrammetric time-series AVHRR Pathfinder datasets in the Krishna River Engineering & Remote Sensing, 56(9):1237–1246. Basin, India, Photogrammetric Engineering and Remote Sensing, Leica, 2010. ERDAS Field Guide, Volume 4, October 2010. 73(10):1029–1040.

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING November 2015 879

11-15 November Peer Reviewed.indd 879 10/21/2015 1:18:39 PM Thenkabail, P.S., M. Schull, and H. Turral, 2005. Ganges and Indus Uppala, D., R.V. Kothapalli, S. Poloju, S.S.V.R. Mullapudi, and V.K. river basin land use/land cover (LULC) and irrigated area Dadhwal, 2015. Rice crop discrimination using single date mapping using continuous streams of MODIS data, Remote RISAT1 hybrid (RH, RV) polarimetric data, Photogrammetric Sensing of Environment, 95(3):317–341. Engineering & Remote Sensing, 81(7):557–563. Thenkabail, P.S., E.A. Enclona, M.S. Ashton, C. Legg, and M.J. De Varlyguin, D., R. Wright, S.J. Goetz, and S.D. Prince, 2001. Advances Dieu, 2004. Hyperion, IKONOS, ALI, and ETM+ sensors in the in land cover classification: A case study from the mid-Atlantic study of African rainforests, Remote Sensing of Environment, region, American Society for Photogrammetry and Remote 90(1):23–43. Sensing, Proceedings of the Annual Conference, St. Louis, Thiruvengadachari, S., and R. Sakthivadivel, 1997. Satellite remote Missouri. sensing for assessment of irrigation system performance: A Velpuri, N.M., P.S. Thenkabail, M.K. Gumma, C.B. Biradar, P. Noojipady, case study in India, Research Report 9, Colombo, Sri Lanka: V. Dheeravath, and L. Yuanjie, 2009. Influence of resolution in International Irrigation Management Institute. irrigated area mapping and area estimations, Photogrammetric Townsend, P.A., 2001. Mapping seasonal flooding in forested Engineering & Remote Sensing, 75(12):1383–1395. wetlands using multi-temporal Radarsat SAR, Photogrammetric Engineering & Remote Sensing, 67(7):857–864.

Delivered by Ingenta IP: 192.168.39.211 On: Mon, 27 Sep 2021 21:14:55 Copyright: American Society for Photogrammetry and Remote Sensing

880 November 2015 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING

11-15 November Peer Reviewed.indd 880 10/21/2015 1:18:39 PM